Plant Biopotential Sensing Based on Generative Adversarial Networks for Environmental Anomaly Detection

Anomaly detection has a wide range of applications in various fields. The anomaly detection of the environment is also one of the important research application areas. In this study, we propose plant sensing using plant potential sensing for anomaly detection in indoor environments. We used an anomaly detection method of adversarial generative networks to identify behavioral actions that capture the anomaly in the environment. For the dataset of plant sensing anomaly detection, we used the electrical signal images and generated the images of the time and frequency domains of the electrical signal as the dataset using the Fourier transform method. In the anomaly detection model, we can not only detect the anomaly of plant sensing signal but also convert the plant sensing signal into frequency-domain signal. The application of plant sensing anomaly detection is compared to surveillance cameras and motion sensors, which can protect private space in monitoring support and can make time-series predictions. It can be applied to detect human movement at home. In the future, it is possible to support caregiving applications for elderly people living alone. Finally, in order to verify the performance of the recognition, through experiments, we have 78.409% and 61.458% anomaly detection accuracy for the anomaly signal.